Tag Archives: cache

Tridion and JMS: removing another SPOF

When you use Tridion’s content delivery API to retrieve dynamic content, resolve links, etcetera, you are well advised to use the built-in caching mechanism. Without it, every page request would result in dozens of queries on your broker database (or dozens of data files being opened, which is just as bad for performance). With caching, most requests can be handled completely from memory, which makes your site a lot faster.

There is one catch to caching: Tridion being a content management system, it is highly likely that the content and links will change from time to time. That is why Tridion offers a notification system, so that your web application will always show content which is up to date and links that lead to the correct destination.

Actually, Tridion offers not one such notification system, but two. Most implementations use the so-called Cache Channel Service. This is a proprietary service which runs as a windows service or a stand-alone java process. Its job is to notify the broker instances when a new version of an item has been published to the broker repository. The Cache Channel Service uses the RMI protocol to communicate between the different virtual machines (which contain the broker instances).

Although this is certainly the most popular notification mechanism, it has some drawbacks:

  • RMI is not a messaging protocol, so things might go wrong when the Cache Channel Service is temporarily down. Fortunately, Tridion has solved this by some clever programming in the broker, but it is still a work-around which makes the CCS slightly unstable at times.
  • RMI uses unpredictable ports, which means that all ports > 1024 must be open between all your machines running a Tridion broker and the one running the CCS.
  • There is no easy way to handle multiple channels (like a staging site, a live site, a mobile site, development environments, etc). You can only achieve this by running more than one CCS instance, which means running multiple processes, each on a different port, or even running the CCS on multiple machines.
  • The CCS is a single point of failure (SPOF), since it cannot be scaled up.

As a solution architect, I find these disadvantages quite unsettling. Especially the last one: removing single points of failure from a solution is what we architects drool over.

Fortunately, SDL has come up with an alternative notification system. Instead of installing the CCS, you can also install a JMS instance. JMS – Java Message Service – is Java’s messaging system. There are many implementations available, including free ones like Apache’s ActiveMQ.

JMS is designed to pass messages from one application to another. These messages are organized in containers called ‘topics’. Applications can subscribe to a topic. Whenever a message becomes available in this topic, it will be passed on to all subscribing applications.

JMS addresses each of the problems of the Cache Channel Service mentioned above:

  • JMS can work over http on a fixed port (typically 61616, but it can be configured to run on any port including 80)
  • Multiple channels can be accommodated on a single JMS instance by simply configuring extra topics
  • Since JMS works on http, it can easily be accessed through a load balancer. This removes the single point of failure from the architecture.
There is a common idea about JMS, and that is that it would be unsuited for a .NET presentation stack. This idea is incorrect. I can guarantee that the solution will work just as fine if your web site is on IIS. After all, the content broker core is always Java, and it is this core that does the communication through JMS.

Where’s the catch then? Well, there isn’t any, other than the lack of awareness of this solution in the Tridion community, and the lack of configuration examples in Tridion’s product documentation. But with some perseverance, the help of this excellent post by Julian Wraith and – if necessary – the assistance of Tridion customer support, it can certainly be done. The end result is a more stable and more robust solution.